Successfully reported this slideshow.
Your SlideShare is downloading. ×

Philosophy and the Social Web

Upcoming SlideShare
WebID and eCommerce
WebID and eCommerce
Loading in …3

Check these out next

1 of 92 Ad

Philosophy and the Social Web

Download to read offline

This presentation was given at the First Philosophy and Web conference in 2010. After describing the issues in the Social Networking space, the presentation moves from a presentation of Web Architecture and it's relation to Philosophy to show how the current problems are philsophical as well as technical and to point out that the way out has been available for all to see. Discussions on this can now take place on the W3C PhiloWeb Community Groups:

This presentation was given at the First Philosophy and Web conference in 2010. After describing the issues in the Social Networking space, the presentation moves from a presentation of Web Architecture and it's relation to Philosophy to show how the current problems are philsophical as well as technical and to point out that the way out has been available for all to see. Discussions on this can now take place on the W3C PhiloWeb Community Groups:


More Related Content

Viewers also liked (16)

Recently uploaded (20)


Editor's Notes

  • Hi My name is Henry Story. This is the talk I gave in French on the 16 October 2010 in Paris at the PhiloWeb conference. I will give it again here again in English, enriched with the feedback from the conference and some new elaborations.

    My background is in both philosophy and engineering. I received a BA in analytic philosophy in London and even started an MPhil, before moving to computing. In 1996 I worked at AltaVista where I developed the Babelfish Machine translation web service until 2001. The past 6 years I was given a lot of freedom at Sun Microsystems, where I developed a deeper knowledge of Web Architecture, Protocols and the Semantic Web and conscientiously blogged on the philosophical/technical issues I confronted, and it is these that I have gathered together here.

  • Tim Berners Lee, who as everyone knows invented the current Web, said a few years ago:
    “The Web is now Philosophical Engineering”.
    Indeed, the parallels between the Web and Philosophy just cannot fail to strike someone with knowledge of both.

  • Let us start with a politico-technical issue - the massive rise of Social Networks in the past decade - that has affected all of us. By focusing on a core issue with these networks I will be able to develop a good understanding of web architecture, and show how by being faithful to it we can solve the issues described. But this will bring us to considerations of issues in the Philosophy of Language and of Mind as seen from this technical perspective.

    The picture here was drawn up around 2006 Ludwig Gatzke. It showed the logos of many top Social Networking Sites at the time.
  • Three years later Meg Picard used Gatzke’s work to to illustrate the change that had taken place in the mean time. The crossed sites here show those that had shut down. These were sites where people had put a lot of enthusiasm and energy to enter information, build communities, and network with people only to find all their work disappear from one day/week to the next, as investment money ran out, sometimes with only a weeks notice for users to remove their data, sometimes with no notice at all.
  • A number of other sites (shown by the green circles above) were bought up by bigger players. Often this meant that the data was still available, but of course this change of management did not come without its own risks, and worries. Would the new management renege on the promises of the founders? Would they be as trustworthy? Would they keep investing? Many artists might have felt very uncomfortable when MySpace got bought up by Murdoch the Media Mogul owner of many, many newspapers around the world including the famous Sun read in England by millions, and always somewhat of an embarrassment in the underground as the page 3 now 5 is nothing else than the picture of a usually large-breasted topless woman, as if to make the reader feel comfortable that this is a man’s paper.
  • So in three years this is the change that took place. Of course a lot of new companies appeared too. The process is ongoing, but presumably slowing down, as people end up moving to the safety of the very very large players.
  • So let us consider an issue that cuts across both winners and losers in this space. Most Social Networks provide their services for free to the end user. They provide a space for people to upload pictures, content, and link with other friends. In the most successful social networks the relationships established between users is what gives them access to certain privileges: access to their friends information, to their friends friends social network, etc...
  • Behind the User Interface given to the user lies a graph of relations such as this one, linking people to things, places, people and content. The confirmed links between users, shown above as two way arrows is what enables access to friends profiles, allows one to comment on their walls, and to leave messages for them. This creates a nice cozy environment where people feel protected from the supervision of outsiders - be it parents or teachers. A place where people feel they can leave messages to all their friends at once, socialise, play games, exchange information, find jobs, etc... The more useful features a site can offer around the graph of relation, and the bigger the social network on that site is, the more valuable it becomes to the user, and the more people will join.
  • But the feeling of coziness is partly a mirage. Whereas every user can only see part of the Social Graph, the network operator can see all the relations combined.
  • A bit of maths here is helpful. Take a set of three people: Allan, Beatrice, and Clive (A, B, C).
    Consider the Power Set of that set, that is the set of all subsets of the set. It contains the empty set
    the set of all singletons, the set of all pairs and the set itself. So the number of groups that can be formed
    with 3 people is 8 as shown.
  • Power sets follow a mathematical law: for a set of size n the power site size is 2 to the power of n.
    It is worth looking at how this grows.
  • The Power Set Size for 100 is 1 nonillion, for 200 it is one hundred sixty octodecillion.
    To get an idea of the size of the Powerset of sets with 1000 members, we need to compare this number to the Google, the number of the famous search engine. A Google is 10 followed by 100 0s. The third number has 300 digits, that is it is a google times a square google. So the space of possibilities we are looking at would be like positioning a Google as one square in a space one google long and one google wide.

    If we imagine 1000 acquaintances listing all the groups they are interested members of then, this is equivalent to them declaring their position in this huge space of possibilities.

    Well those are some back of the envelope calculations that would require some further thought.

  • The current social networks act somewhat similarly to Bentham’s Panopticon.
  • Bentham devised the Panopticon as an architecture for improving the quality of life for prisoners. In order to justify the improvement in the quality of life he sold its economic advantages. His argument was that it required only very few guards: by placing them in the central tower they could observer the prisoners placed in circle around them. The prisoners had only a limited view. The guards could on the other hand could see without being seen. The prisoners never being sure they would be seen had to integrate the laws of the guards even when they were not being looked at.
  • But is this really a good parallel to Social Networks. Nobody is forcing people to stay in them. There are no guards at the doors. Everyone can leave.

    But what is the cost of leaving? One can now leave a Social Network with all one’s possession (photos, entries, friend contacts, etc...) without trouble, but when one leaves one can no longer interact with them. By leaving the current social networks one looses one’s automated relationships to one’s friends, but also to their friends, and to the network as a whole.
  • And what advantage would one gain if it is just to leave for another, perhaps smaller, less effective network?

    Each of these networks is known as a Data Silo: data is locked inside unable to connect out. Each network one moves to requires one to re-establish all one’s connections, to re-enter all information. If one opens too many such networks one is bound to fall down under the work of synchronising information across them.

    Does it really have to be like this?
  • Consider the telephone network for example.
  • Calling across network operators does not seem to be a problem there.
  • It may have been at one point though. But the following reasoning can show us the pressures that are thought to have been brought to bear in that space.

    Consider a time where only two telephones were connected by one line. The value of the line would be completely dependent there on the value those two people laid on their ability to communicate with each other remotely.

    Consider now a network of 5 telephones. With 5 telephones the value of the telephone for each user is the value he puts on communicating remotely with any of 4 other people. The value of the network is the combined number of such connections, namely 10.

    With 12 people in the network the value of the network is 11 for each user and so 66 for the whole network.

  • Imagine now that a country such as Germany and France were totally disconnected islands. The value of the German network would be according to current population 2 080 million squared (quadrillion), the value of the French network would be 3300 quadrillion links. By laying one cable linking the German and French network, both the German and French networks gained millions of users, but more interestingly the combined value of the network is double the value the sum of the two.

    This gives an indication of the value advantage for any telecom operator to be part of a global telecommunications space. It used to be argued that most people would find nearly no value in connecting out of their local community, and so that these figures should not be taken too far. But more and more we are finding that those links, once in place do get used... When spaces of possibilities open up, organisations develop to start making use of them.
  • In any case one can now call people from any operator to any other operator, without ever really knowing what operator one’s contact is on.
  • All one needs to know is their telephone number, a globally unique identifier for their phone. In some countries one can even switch operator and keep the same number.
  • The same is true of e-mail.

    In the early 80ies, when universities were starting to connect with each other, the value of e-mail was still very limited. Even academics who could with minimal effort get email accounts often found the inconvenience of learning new tools to participate in this network not to be worth their while.
  • But now with email you can send messages to the whole world, whatever organisation your contact belongs to, wherever they are.

    Again e-mails rely on globally unique identifiers, identifiers of inboxes that are easy to write down at a party and type in a message. Nobody needs to know which software runs the resulting system, or who runs it. E-mail was designed for cross organisational communication.
  • Now Consider the Web, you can buy your own domain name ( mine is ) and so can your friends...

  • Imagine all of us get a computer and name each of our boxes using DNS (Domain Name System). The computers could be anywhere: in the cloud, in your basement (of if you are Swiss in your bunker), in a pocket sized personal computer devices, in your cell phone, .... whatever. We don’t care. We just draw them as boxes.
  • On each of these you can set up your own web server which then allows others to access your web site with a URL such as mine

    You then can place web pages on each of these servers. Any number of them. They can then be linked.
  • Here every document has a glovally unique name. Some of them are listed here.

  • It is worth pausing here and looking a little bit more carefully at what URI are. These are introduced in RFC3986 as “a compact sequence of characters that identifies an abstract or physical resource”, where the structure of the sequence is the further defined in the RFC. URIs such as a well respected center left French newspaper, are thought to be refering then to a resource, shown as the oval in the diagram, which in this case is usually understood to be “The Front Page of Le Monde”. This thing it refers to is and always will be the same thing, which can cause some confusion since the front page of newspapers changes daily if not hourly - they would not be interesting otherwise.
    But this is not that different from a name naming a person. People are a cell large at conception and then grow into fishlike beings before appearing in the world, crawling about, standing upright to walk to work, where they wear themselves out over time into old age and wisdom. The standard way to conceive this is to understand that a person is (at least) a four dimensional space-time continuum composed of time-slices which are the parts that at any moment we see. Similarly the front page of Le Monde has variants, which themselves can - even should - have their own URIs. Giving URIs to variants would allow one to consult front pages in the past and refer to them easily.
    But the difference between the front page of Le Monde and a human is that the the latter is a natural kind whose evolution is determined causally by biology, whereas the identity of the front page of Le Monde is determined by that organisation. They could change it at any point to refer to something else perhaps to an description of a kitchen sink making the meaning of the URI thereby very unintuitive, a bit like the well known concept grue (green before 2000 and blue thereafter) well known to philosophers. Is the meaning of a URI then completely arbitrary?

  • Even though each individual site owner can determine the representation he puts up on the Web at a URL the linking to that resource is not under his control. The purpose after all of publishing information is to make it available, and the best way to make it available is to make it linkable, to grow its position in the web of quality information. The easier and more predictable the entity named by a URI is for the users and linkers to that resource, the happier they will be, as it will allow them to build an information ecosystem on top of it.
    If the front page of Le Monde were to one day just show information for a kitchen sink the users of that site would be clearly confused, the social community linking to it would feel deceived, and if that persisted long enough would be forced to remove their links. The restrictions on meaning are social. Each information owner is enriched by every link, especially by quality links to his site. To break the expectations of the linkers is to loose social trust, and to loose a position in the quality information space, which as far as information goes, is the most important value to maintain. To not maintain the social trust values is just as in real life to loose one’s word, and so soon to be unable to be heard.
    But then the question is how do people know what the meaning of a URL is? At present that is usually something that can be guessed by the human readable description of the document. The Front page of a newspaper is something everybody understands to be different from one day to the next. Same with a page for currency converters. A blog entry on the other hand has the notion of a permalink, the link where one expects the blog entry to remain for a long time, with very little changes, the one that one should link to if one wants to speak of some text.
    As we saw before the value of a network builds up over time, and the value grows with the number and quality of the links. Many sites such as Le Monde break these expectations, by for example making it impossible to link to articles over time. Le Monde seems to for example move newly published articles to a different “archive URL” which is impossible to find from the original publication URL. This means that they interact very badly with the blogosphere, search engines, people sending each other mail about an article. In fact their behavior is an incentive to copying information instead of linking to them. I was told that recently Le Monde even changed the meaning of all their URLs, breaking all sites that linked to them. No wonder they need to go for help to the French state to survive. Technical issues are cultural issues, and to want to ignore that is an expensive mistake.

  • Linking across sites is a trust issue that fragilises the publisher/speaker/agent (to use a term coined by the psychoanalyst/painter Bracha Ettinger) by making her dependent on the behaviour of other players in the space, but also strengthens her if the trust is rewarded. This is the same in any social endeavour from playing music, to building bridges, to marriage, to shopping, to electing presidents, as John Searle emphasises so well in his latest book “Making the Social World, The Structure of Human Civilisation”. The Web as a communication mechanism, can be no other than social, since language and civilisation itself is thus.
    Speaking of Searle. He is the author of the famous “Chinese Room experiment” where he imagines himself in a room receiving symbols from the outside and manipulating it by following a set of purely syntactical rules, by following a huge rule book, in such a way that the reply he returns would count as a good answer in Chinese. As he points out in his online 2010 lectures, the key point is that Turing Machines are purely syntactical machines. But there is something striking here: the Web is a set of interconnected machines, Turing machines perhaps? just processing symbols. Yet the web does clearly have a minimal semantics: a browser can fetch a document placed on a server on the other side of the world; documents link to each other; ...
    It could be argued, and I think Searle would argue, that it is we who give those machines their semantics, by tying them into large telecommunication structures, that follow carefully established protocols, which form expectations on which actors start basing themselves. It is quite staggering when we think of just how big some of these human built structures are, how much knowledge and coordination they involve. It is clearly with the purpose of referring to things that URLs are built, this is how web developers think of them when using them, this is what people think when clicking on a link. That behind this the turing machines just shift bits and bytes while reductionistically true, may also make things suddenly seem very mysterious. Similarly explaining human behaviour in terms of Neuron firing may at times be useful, but mostly not that helpful in everyday life.
  • John Searle book, builds on his work on Speech Acts. The human world he argues there is built coordination which is only possible at such a scale because of language and its use. As we are looking at the web we can perhaps see something similar here. At least it may help us open up some new ways of thinking of it.
    First we can think of all computers on the Web as agents in so far as they can issue and respond to requests. The interfaces used on the Web are particularly simple ones, which is what enabled this system to grow so quickly to the scale it has. The main Speech/Document act is the GET request, whereby one computer agent asks another one for a REPRESENTATION of a resource.

    Consider for example a Browser in the illustration above (also known as a User Agent) requesting the web page . To do this the User Agent opens up a TCP connection to port 80 of the machine and issues the request “GET / HTTP 1.1” followed usually by some further metadata describing the capabilities of the browser. The Server running an Apache httpd daemon perhaps - there is a lot of other software implementing the protocol - looks up the file in the file system, or generates one on the file from data in a relational database and returns the HyperText Markup file preceded by a few headers of which the most important perhaps is the mime type of the content, which enables the browser to select both the correct parser for it and to determine how to display or use the information. The html usually has pointers to further embedded images which will then be fetched the same way. Even a document on the Web is not just one file, but a graph of interlinked documents!

  • There is nothing in the HTTP protocol that specifies that the client has to be a laptop and the server a machine without a User Interface. The Client/Server distinctions are just roles that computer agents play in the protocol. To make this clear I have here replaced the laptop with a box, which could be the same laptop, or another server. Search engines after all follow exactly the same protocol to crawl the web for information before handing their discoveries to indexing software.
  • Indeed you can yourself try being the client. If you have a terminal on your computer and are connected to the internet you just need type at the command prompt the command shown in red above. First type “telnet 80” and press return. Once the connection is established type “GET / HTTP 1.1” and two carriage returns. The server will respond with a header and the XML shown and will close the connection.
  • One very interesting feature of the way HTTP was designed is that it is possible to cache the representations returned, either on the server, or in an intermediary cache server, or indeed on the client itself. Caching on the client is very useful for speeding up browser behavior as that can avoid them having to fetch the same image again and again. Middle caches were very useful initially in order to speed up transatlantic connection, or to allow servers to poole resources. Many Internet Service Providers can also speed up the response time to their users by caching videos or films. This is possible because HTTP is designed for stateless requests. That is when requesting a remote resource the server need not (and as far as possible should not) keep track of any state of the client it is serving the resource to. When servers stick to this architectural constraint it reduces the load needed on the server and the web scales much better.
  • Not only is it possible to have such a central cache, but it could also be possible to have a p2p caching system, where you ask your friend for a representation on their friends web site. The same server could then play the role of a server a client or even a cache. I don’t know of anyone who has developed this but it is an intriguing possibility.

    Would it then also be possible to ask for representations from a certain date? Perhaps one could ask for the representation someone had for the front page of Le Monde on September 11, 2001. This would require further investigation.

    Like any language it is possible and indeed to refine the protocols when the needs are clear enough. A well engineered system is one that permits such refinement without breaking the existing implementations. Hence the very strong emphasis in the Web on minimality of specifications, openness to future changes and declarative specs.
  • So an interesting project in the Philosophy of the Web would be to work out what the HTTP “Speech acts” are, how these map to John Searle’s categories of Assertives (this is how the world is), Directives (“do this”), (Declaratives “The Session is now closed” as said by the chairman, “Guilty” as said by the judge), Commissives, (where the utterer commits himself to something - reserving a ticket for example), Expressives (Emotional feedback). These may not at all be mappable to these three verbs, but understanding why this is so, and what their role may be, what else would need to be added, could be very helpful.
  • A lot of the above is a summary of Roy Fielding’s famous Doctoral Dissertation “Architectural Styles and the Design of Network-based Software Architectures”, where he describes the type of architecture style of the web, what the properties of this architectural style are, and so explains how this largest and most successful engineering feat was possible at all.
    To summarise the HTTP protocol is a client server protocol, where the client and server roles can be played by any computer. The protocol aims to be Stateless, so that as far as possible representations returned are not dependent on the identity of the agent accessing the pages, simplifying the roled of the server but in particular making it possible to cache information. The system is a layered system with a uniform interface.
  • The Web as we have described it up to now has consisted only of URIs referring to resources that have been some form of what is known as Information Resource - also known more commonly as a Document. In the case of URI referring to a document, the sense of the document is given by interpreting the document. So the Sense/Reference distinction that we get from Gottlob Frege, who revolutionised our understanding of Logic at the end of the 19th century, and on which Bertrand Russel, Wittgenstein, the Vienna circle and essentially the whole of the 20th century analytic tradition of philosophy based their work, these two relations have up to know been co-extensive.
  • The Semantic Web is one step forward on what is known as the Web of Documents, in that it permits the Sense/Reference distinction to be made and used by taking advantage of a feature of the URI that has been hidden from view until now: the fragment identifier. RFC 3986, which defines the URI has this to say: “The fragment identifier component of a URI allows indirect identification of a secondary resource by reference to a primary resource and additional identifying information. The identified secondary resource may be some portion or subset of the primary resource, some view on representations of the primary resource, or some other resource defined or described by those representations”. It is this later feature of fragment identifiers that is used by the semantic web: The reference of a URI with a fragment identifier can be defined or described by the representation returned by the document referred to by the URL without the fragment identifier. In the case illustrated here, the URL “” refers to me via the meaning of the document

  • So now that we have the reference relation, we can explain what is semantic about the Semantic Web. Semantics is the relation that ties URIs to their referent, the relation that ties strings to things we might say, or words to the world. Syntax on the other hand is the way names for things or properties can be combined to produce what are known as well formed sentences or documents.
    In the case of the semantic web we just need to imagine the world composed of things and binary relations between things. Here we have in the Photo which I took a few years back in Berlin two people Richard and Anja, related by one relation. This relation is the relation named by the URI .
    So the top layer of three successive URIs forms an RDF sentence. These are just strings. The relation in the world is what is satisfies the sentence according to this interpretation. The relation between the words and the world (again a binary relation!) is what semantics is about.
    This may seem very weird btw. We know that computers are very good at manipulating symbols. How do they get this relation to the world? How do they get Semantics? Do they really need this relation? Why this extra layer that seems to parallel the syntactic one? I will try to disentangle the magic here. But don’t let it entangle you. Human beings find it quite easy to distinguish words and the world, and understand that relation very well. After all the string for Anja in the picture is 47 characters long and as opposed to Anja never had a mother, and never saw Berlin.
  • This reminds me of the statement that Tim Berners Lee made a few years ago at conference when a debate about what the Web was, started to heat up, with Browser focused engineers claiming the Web was anything that ran in the browser, and others explaining that this meant pretty much anything, since with extensions to the browser any data structure available on the internet could be used to produce pages. Tim Berners Lee settled the discussion somewhat by proposing two definitions of the web, a short and a long one. I can’t remember the long one, but the short I remember clearly: “The Web is a mapping from URIs onto meaning”.
    But this could make things even more mysterious. What are meanings?
  • Meaning is in fact formally defined by the RDF Semantics spec as follows, but it holds just as well for sentences in
    any language:
    “The basic intuition of model-theoretic semantics is that asserting a sentence makes a claim about the world: it is another way of saying that the world is, in fact, so arranged as to be an interpretation which makes the sentence true. In other words, an assertion amounts to stating a constraint on the possible ways the world might be. Notice that there is no presumption here that any assertion contains enough information to specify a single unique interpretation. It is usually impossible to assert enough in any language to completely constrain the interpretations to a single possible world, so there is no such thing as 'the' unique interpretation of an RDF graph. In general, the larger an RDF graph is - the more it says about the world - then the smaller the set of interpretations that an assertion of the graph allows to be true - the fewer the ways the world could be, while making the asserted graph true of it.”
    In short Meaning is a set of possible worlds, the set that make the document true under the interpretation for the language the document is written in. The bigger the graph of information the clearer the picture of the world. For example a large Social Networking site with 500 million users each with 10 of thousands of relations, some entered explicity others discernible through behavior, all interlinked will have a much more focused description of the actual world than any of the participants in the network has.
  • So to summarise the URI above that refers to me, maps to a document whose interpretation sets a restriction on the set of possible worlds which can have me as a referent. In this case perhaps there won’t be enough information in the graph to distinguish me from many other people. But we’ll see how we can get to a definite description that would tie that description down later.
  • With this in place it now becomes possible to link not just documents, but also to related people via those documents into machine readable social networks distributed across servers. Juliet can publish information about herself as well as relationships of various kinds (friendship, kinship, etc) that she has with other people across the web. These in turn can confirm those relationships by publishing profiles linking back to Juliet’s via what we can call her WebID.

  • So let us steer back from the heights of theory to something practical. A few years ago I developed a Semantic Address Book, which could also be called a Hyper Address Browser, as it allows one to browse the distributed social networks the Way a browser allows one to browse the hyper text web.
    But the best thing to do is just to have a look at this video as it will make it much easier to understand what we are doing. This is a proof of concept software, which is the equivalent in engineering to the Philosophical Thought Experiment.
    I cannot include the video in the slidecast here directly at present, so I am linking to it. Please click on the link and watch the video before continuing.
  • Let us consider now what the network effect of the Semantic Web looks like, as compared to the telephone network, or the network effect of static documents.

    Consider the simplest case which is two resources Romeo and Juliet. On the Semantic Web there are any huge number of relationships that can relate these two people. Here we show just two of these: a relationship named “knows” and one named “oppositeSex”. I leave it up to the reader to come up with more.

    Consider next the case of a graph referring to five people. We don’t show here all the relationships we could possibly draw between them, as we could easily end up with as many relations as there are sentences in Shakespeare's play. I did want to highlight a new type of relationship we have not encountered before: the inferred relationship. The foaf relationship here stands for the friend of a friend relationship and is true just in case two people are related by an indirect “knows” relationship. So since :juliet knows :FriarL and he knows Montague there is a foaf relationship between :Juliet and Montague. The foaf relationship between :Romeo and :MmeCapulet is explained by the fact that the loves relationship between :romeo and :juliet implies a “knows” relationship just as the :mother relationship between Juliet and :MmeCapulet does too.

    It is clear therefore that the connectivity between semantic networks grows a lot faster with any piece of information that the telephone network. Somehow this should be useful for overcoming the fragility of the relationships described as compared to the solidity of the “can make calls” relations in the telephone network.

  • Earlier I showed how a Resource can point to a Person. But the Semantic Web very importantly also allows it to refer to abstract entities such as classes of things. Here for example we show how the foaf:Person relation refers to the set of people. The document that is returned by the URI explains in a comment in English what that URI refers to: namely the class of people. It is this human readable text that is so easily accessible (one click away on the internet) that makes it possible for programmers developing software such as the Address Book I showed earlier to write software that will fit the expectations of other people publishing data using this vocabulary. It is the natural language text that ties the machine readable relations to semantics. But most importantly of all it is the ease with which the documentation for this word can be found that makes it easy for people around the globe to come to some consensus on the usage of the term.

    Notice btw. that that relation does not have a fragment identifier and yet does not refer to a document. That is because there is another trick for tying URIs to things via documents and that is by using an HTTP 303 redirect mechanism where the server on receiving a request for that resource will say essentially “I can’t server up a relation, that’s not a document, but go see over there”. You can see that this happens if you type that URI in your browser address bar: it redirect you as shown

  • ... here to the specification. If you had asked for the document using a tool that preferred an RDF/XML representation the server would have redirected you to the purely machine readable format. Here we are looking at the foaf:knows relation. You can see it is defined quite fuzzily. No problem with that.

    “We take a broad view of ‘knows’, but do require some form of reciprocated interaction” it says. “Since social attitudes and conventions on this topic vary greatly between communities, counties and cultures, it is not appropriate for FOAF to be overtly specific here.”

    Good vocabularies if they want to spread should try to avoid making uncessary distinctions, as those will very likely get lost in deployment. Dan Brickley and Libby Miller who put this spec together did a very good job on getting this balance right.
  • So let us now consider how the Address Book works in a bit more detail. Let us imagine we have two servers on the web one by Tim Bray one of the authors of the RDF/XML spec, a famous blogger, and semantic web skeptic, and one by me. Let us imagine that one day Tim Bray will give up his skepticism and just use the semantic web by publishing his foaf profile in some document such as the one shown here, which is a readable non XML notation knows as Turtle a subset of N3. The two documents describe two people and link them by the knows relation. Clearly there are no arrows going from one server to the next. The link is made by the use of URIs of course.
  • But there is no harm in picturing the documents as two graphs that link to each other as shown here. Each person is described by his relationship to a few other resources, be it an openid a telephone number or a name. And each graph links each primary topic of the graph - that is Tim Bray on the left - to the primary topic of the graph of the other graph via the foaf:knows relation.
  • What the Semantic Address Book does is to fetch those two graphs using the HTTP protocol, which it then stores in a semantic database (knows as a quad store) whilst adding metadata on the graph relations themselves. That is the semantic database makes sure it can keep track of where and when it found the information.
  • The Address Book can then merge the information from the two graphs to produce a new merged graph. Where one could not ask what the name of the friends of the primary topic were in the original mini graphs, one can do this now. Here one can ask for example what Henry’s friends’ names are using a query language such as SPARQL. The answer here will be “Tim”. By merging the graph we deepen our ability to query data, we can explore a larger network and we can discover new relations that were not available in either of them alone. The Power of the Semantic web lies exactly here: it is one could say the mathematics or logic of information merging. Everything in the semantic web is designed to make this easy: from the simple relational data structure, to the open world assumption, to the URI reference distinction.
  • This merging of information can help us understand even more clearly the importance of reference on the semantic web from a computational perspective.
    Imagine for example that our Address Book had found the following two graphs on the web, each of them making an identity claim. The owl:sameAs relation is this identity relation. From the law of indiscernibility of identity it follows that the object, however named must have all the properties truly asserted of it.
  • The merged graph therefore can merge the relations from both graphs onto the object referred to by the two URIs. Clearly the two URIs are distinguishable. They have different number of characters for one. Their referent is therefore clearly different from the URI. It is the referent of the names that is what gets merged.

    The concept of reference is therefore essential to making merging of information possible. And since merging information is what allows the powerful network effects to take place, it is an essential tool of thought.
  • This is a point P.F.Strawson put as early as 1975, whilst explaining where the informational value of identity statements with different names came from. As we have just seen it is the merging of information that provides this value.
  • So as I was happily showing my Address Book to people in preparation for the JavaOne conference in 2008, I again and again got the same praise and criticism: “this is just great” people would say, “but is there a way to protect the information”. Most people don’t want everyone to know what their Social Network is. Protecting their information is important for many reasons, be they business reasons - sales people are very careful about their good leads - to people with children who don’t want just anyone to know where they are and what school they are going to, to people wanting to protect their property: there is nothing more useful for a thief than a map of people on holidays.
  • But if we are going to have 100s of friends distributed around the web, each with their own servers with every resource protected and accessible only to authenticate people

  • then we are really going to have to solve the login problem. What is the problem here? Well consider the following each server will need to know who is connecting in order to decide whether or not to server the resource. If you are a friend of a friend, my personal Social Web CMS (Content Management Server), should be able to server you contact information, friend relations, and give you access to party and other resources. Family and businesses might get access to other resources. But in order to decide my server needs to be able to know who is connecting.
    And if we use username/passwords to do this, the most to authentication method on the web, then we won’t even get off the ground.

  • I will show next how one can use the Semantic Web and the theory of reference to get provable global identity to solve this problem cleanly and simply. At the same time we will see how this issue has been there for the taking for 10 years. When one has a solution to a problem as important as this that is as simple as the one to be demonstrated, and that it remains invisible to all engineers even though it is globally available, this demonstrates clearly that the issue is no longer in the technology but in the ability of engineers to step out of a certain deeply ingrained way of looking at the world. An issue that demands a philosophical investigation.
  • Even though username/passwords are widely understood, they have a number of very big problems associated with them. Biggest of all is that they are incredibly tedious for end users to set up and as a result end up providing no security at all. Because they are tedious to remember, people tend to use the same password on every site. So in a network of 500 contacts most people will use the same password more than once. This means that one just needs one rogue friend, and he can impersonate one on all one’s friends servers. In that case one may as well not have any passwords at all. Furthermore username password schemes require one to create new accounts on each server and so also to loose one’s global linkable identity. So we’d be back to the problem with current social networks: each of our friends Social CMS would behave very similar to a data silo, except that these would be a lot smaller and a lot less useful.
  • Let us now consider one very widely known solution to the global authenticaiton solution: known as OpenID.

    Essentially by remembering the short URL for your home page you can authenticate yourself at a web site that you have never been to before, without that web site ever knowing your password.

    Even though OpenId is widely available, and much talked about, a lot of people still do not have a good idea of what it is. If you are in that situation, the following very short video will show you how it works in action. That should be enough for the purposes of this talk. (Though you can skip this part without too much loss)

  • As one can see from the screen cast and from this UML Sequence diagram the protocol is still somewhat too complex. In particular it is more complicated than the username/password scheme. Even though it does not require the person logging in to a service to remember a password, it still requires him to go through a number of steps. There are 4 Actors in this UML diagram Romeo who wants to login to Juliet’s server, Juliet’s server, Romeo’s Server and the OpenID Identity Provider. There are also 8 HTTPS redirects needed to make this protocol secure. And finally it does not tie correctly into Web Architecture, which is the biggest problem, because as we have seen it is web architecture that is the reason why the web is scalable as it is and why the network effect works so well.
  • In contrast, The WebID protocol shown here is a lot simpler.
    Most obviously of all It requires only one connection over and above the initial request, and the result of that request is cacheable, that is there is no need for the server to make that request for every connection.
    This protocol relies on cryptographic tools that have been implemented in all browsers since 1997 when Netscape first shipped SSL version 3 in their browsers, thereby securing access to server and enabling the .com revolution. Without secure connection to a server there could be no e-commerce as it would have just simply been way to risky to send credit cards over unprotected networks. SSL also gave some form of guarantee that the client had connected to the correct server.

    The SSL stack (now know as TLS - Transport Layer Security, which is the standard version of Netscape’s partly proprietary protocol), also enables the server to ask the client for its certificate. This is what we are using and what we can see in action next
  • The Video Linked to here just shows three very simple things, that are so simple though that a linguistic description can only fail to get the point across. It shows:
    1. That it can be extremely easy to create a WebID
    2. That you can then use it to login to a site you have never been to before in only 1 click
    (The most amazing thing here is that it is integrated in the browsers you are already using)
    3. Finally that this same procedure applies to any otherbrowser just as easily. We show it working with Google Chrome, Firefox and Opera.
  • So let us go over what we saw again. After creating a client side certificate, one signed by Romeo’s own Social Web CMS, he connects to Juliet’s server in (2) above. Her CMS asks the person at the end of the connection for his client certificate. Romeo selects his personal one and sends it over with a proof that he owns the private key matching the public key in the certificate. Juliet’s CMS sees that the certificate claims that the requestor is the referent of the WebID, so she goes and fetches its meaning from the web. Romeo’s CMS responds to that request that confirms that the referent of his WebID is anyone who knows the private key of the given public key. At this point Juliet’s server know that the agent at the end of the request is the referent of the WebID URI. All she needs to do next is to check in 5 what privileges that agent has. If she met Romeo before he may already be in her list of acquaintances. Perhaps Romeo is not known to her directly but to a friend of hers. This may be enough of a relation for her to release the resource.

  • So why has this not been solved or seen before? The answer is not that distant from the reason some people find it difficult to see the old woman in this picture once they see the old woman or vice-versa. Once one is stuck in a certain way of perceiving the world it can be very difficult to shift one’s outlook. One needs to clearly be shown and repeatedly so how to move from one way of perceiving it to the next.
    From travelling the world and giving talks on this over the past year and a half, I can see that this is very much related to what is known as an conceptual shift or paradigm shift. Even though many people use the web every day, and even though many developers develop on it, it is not clear that after 15 years the distributed philosophy of the web has yet had a deep impact on people. As we saw large newspapers still make basic but extremely expensive mistakes by changing the meaning of their URLs. The security industry that probably deals very little on the open internet is even more distant to the idea of distributed peer to peer security. Security tends to be only taken seriously in strongly hierarchical organisation, or at least that is where the serious security money is.
  • The security industry tends to work on the paradigm of the passport. A passport is a piece of paper that is difficult to forge. It is a claim by some trusted third party that the person depicted has the properties described. In this case it is a passport issued by the United Kingdom Government of her Majesty the Queen. A border officer will inspect the paper for proof of tampering to verify the authenticity - the author - of the statement. Having verified that the message is from a trusted party he should check that the person described is the person who gave him the passport. We can think by the way of the picture here as my public key, and my face as the matching private key. So by looking at the my face he can see that the claim made by Her Majesty’s government is indeed about me. He will then check if I have access rights to the border zone and if so let me through.
  • What we seem to be proposing is something rather odd. We are proposing that when going to the passport office instead of showing the officer a passport, we show him a self signed certificate with a WebID. The officer can indeed see that the picture matches my face, but what is he to make of the URL as an identifier? The problem is that the sense of humour of border guards is not something worth testing.

    But let us assume we have a rational border guard with a lot of patience and a philosophical bent. Perhaps he would find it less troubling if I put a government issued URL there, especially if he could check it up on his computer connected to the internet. That computer could show a picture of me, and if my face matches the picture on the internet, we would not be that far from having a virtual passport.
    If one accepts this step, then we are nearly there. The reason he found me on his computer would be because I went to the government offices which issued me with a WebID.
    Now we are just claiming we should do the same, but without a central server, in a p2p web of trust manner. Instead of giving people we meet or know a government issued URL or a complex public key, we can give them at parties or during meetings a much more easy to transmit global identifier: an email tied via the WebFinger protocol to the public profile, a home page linking to the WebID or something of that nature.
  • So to go back to how things are currently. Security officers tend to think in terms of a hierarchical Public Key Infrastructure, a bit like the border guards do. As this is how server authentication works at present, there has been no reason for them to reconsider their view.

    Let me describe how the current TLS server based security framework on the web functions. It is built around a number of well known Certificate Authorities, that sign public keys sent by owners of servers, as shown here. This can be done in a chained manner, with one CA signing an intermediate CA which ends up signing the root CAs.

    The owners of the leaf certificate place the private key that goes with it on their web server which can then use the private key to sign requests made by clients during a TLS connection. Those clients (ie any of a number of browsers on your desktop) can trust the Certificate sent by the server because it was signed by one of the Root CAs certificates stored in their keychain.

    Getting one of those public keys into all the volume browsers is a very lucrative business. It is what made Mark Shuttleworth an overnight celebrity when he sold Thawte to Verisign for 3.5 billion in December 1999. He used this money then to be the second self funded private citizen to go in space, and has been investing heavily in funding the Open Source Linux distribution Ubuntu.

    So why could one make so much money there? Simply because every single business that wanted secure transactions on the internet and wanted their end users to see a secure padlock in the browser needed to pay a certain amount of money per year to one of these CAs to authenticate their business.
  • When Security people think of a Peer to Peer authentication they then automatically think of an older technology PGP. This was also the idea of the Web of Trust.

    PGP requires each user to create a self signed certificate. He can then get this signed by friends or others at what is known as key signing parties. Here we show three certificates owned by Jane, Bruno and Henry, with Henry having signed Bruno’s certificate, Bruno having signed Henry and Jane’s and Jane having signed Brunos. Each could then upload these keys at public key servers.

    A software agent could then lookup some of the public key repositories to find if anyone attempting to communicate with him had his key signed by someone known to him. In the above example for example I could receive mail signed by Jane and feel comfortable that she was indeed someone known to a friend of mine. But trust is not transitive, and key signing requires a lot of conceptual apparatus that is now easy to explain. It is not point and click as we saw WebID is.
  • The advantage of WebID is that the certificates can be generated in one click. This contains minimal information resulting in minimal information leakage. The rich information about the end user can then be placed at a the potentially protected public profile document or linked to from there. This means that there is no need to change information inside the certificate if the user decides to add a new friend, change his address, telephone number etc.... It is MUCH easier to produce good user interfaces to automate changes in such information on a Social Web CMS, than it is to propagate changes to signed certificates.
  • We can now look at the computational cognitive mechanism for identification of things that are not documents.

    Remember that we started off with URIs referring to Documents and showed how the Web also allows them to be retrieved via what we called HTTP Document acts. We then saw how URIs could also refer to things in the world via the description or sense stated in that document, but we noted that this type of referring is often done through relations to natural language which programmers can then use to build software. Machines seem to still be acting mostly as manipulators of syntactic tokens.

    With WebID we are now enabling a machine to identify by reference a real three dimensional Agent (me in this case) using a cryptographic proof.

    So imagine me connecting with my WebID to a server. The server after establishing that I have the private key of the public key my browser sent it, can GET the sense of the WebID which states that the referent of the URI is whoever can unlock the public key. Since it knows that I have the private key, it knows that from its point of view, the entity at the end of the TLS connection (me) is the referent of the WebID. From there it can choose decide or not to send me the information requested.

    The cryptographic procedure is a cognitive mechanism implemented by the server to identify agents making requests with global names.
  • Just as a quick technical note: public and private keys are just sets of numbers. So they could be copied onto other computers without the user noticing it. Luckily as we are building on very well established technologies, it is possible to place the private keys on inaccessible portions of hardware devices such as USB sticks, so that such copying could not take place. To get access to someone’s private key one would need to steal their physical card, which is something that people can be expected to notice.
  • To summarise the talk up to now. We have looked at the Technico/Political Issues of current Social Networks. How the value of these huge network is correlated to the size of the information graph they build. The value of this graph can be determined in part by the huge investments that have been poured into those companies.
    The Web we showed is architected as a peer to peer network, in which the same network effect is at play in the form of links between pages. The meaning of resources over time is constrained by the web of pages in which they are located and the social expectations these create.
    From the philosophy of language point of view we saw how the sense/reference distinction is enabling much finer linking to take place, allowing not just linking of documents but of objects too. We saw how this can be tied to a proof procedure for helping web servers identify people on the web in a completely distributed fashion too. Let us move from this mini cognitive ability a little deeper into the philosophy of mind.
  • It is really interesting to notice how what we find on the web finds echo in literature and elsewhere. The Meaning of a URI we saw is defined at another document, which can link to yet further documents in a web of meaning. Each document links to other documents which have things written in natural language which themselves are explained in dictionaries, where each entry sends to yet another entry.

    That meaning forms such a web leads to the evidence that understanding also forms a web. This is very evident when reading Joyce’s Ulysses for example. Try reading the first chapter unaided in English. Don’t use too many tools to aid you in your understanding, just read attentively. Then open the new comic strip “Ulysses Seen” and see how much you did understand. Once you see what he wrote about, you will notice passages that you misunderstood in the book, but on re-reading you may find that it was there clearly in black and white. Now every page of the comic has a page of notes attached which are very much worth reading as they will inevitably deepen your understanding and change your reading of what the book is about and what is going on. But those notes send you on to yet other links on the web - old Irish songs, famous poems such as Walt Whitman’s “Song of myself”, philosophical discussions in the early Christian Church - and each of those texts of course being written in response to yet other thinkers to understand them really well you may be tempted to read those writers too. Every such deep reference in Joyce makes you aware of the author’s knowledge which can itself force a re-reading of what the chapter is about as well as what the characters are saying. This is what led Derrida in the late 1980ies to imagine a “Joyce Machine” which would be a computer that could link meanings together... Not so unlike what the web is.
    But of course one’s reading of the text only deepens at each reading. We are not emptily looking for meaning all along, in a never ending quest to find the first meaning. There is some minimal grasp that we have on each sentence, that can be furthered by references to dictionaries or encyclopaedias or pointers to songs online. At many points the novel latches on to some cognitive skills that we do have, enabling us to get a grip on this text as it walks us through the whole of European culture by portraying the thoughts and events in one day in the life of an artist in Dublin.
  • Reading Joyce is a good way to greatly deepen one’s own understanding of other minds. But to get to the essential issues in the philosophy of Mind, a good place to start is the place where this is lacking as in the case of autistic children. Whereas healthy children learn by the age of 4 to distinguish what they believe to be the case from what others believe to be the case. Autistic children have trouble with this: they are said to have failed to develop a theory of mind.

    Children are tested through a muppet show such as the one illustrated here. This is how the story goes.
    Two girls Sally and Ann are on stage with as only props a basket and a box. Sally places a ball into the basket then leaves. While she is out Ann takes the ball from the basket and places it into the box. When Sally comes she says she wants to play with the ball. The children are asked where she will look for the ball. Healthy children will have concluded that since Sally did not see Anne move the ball from the basket to the box, she will look for the ball in the basket. Autistic children will tend to state that she thinks the ball is where it actually is - in the box. Of course this is not conclusive. The children who pointed to the box may have misunderstood the question, or they may have reasoned that Sally having noticed before leaving as they did that the ball made a bump in the basket, that on returning she would clearly have noticed that the bump no longer was there and so have concluded that it must be in the box.

  • Now if one tries to simulate the muppet show with any of the widely available programming languages one will have exactly the same problem as Autistic children do. This is because industrial programming languages are designed to build objects - user interfaces, editors, pipes, etc... - and rarely come with any tools to simulate beliefs.
    So consider a naive implementation of the play. We could create objects for all the artefacts - ball, basket and box - and for Sally and Anne. To have them have belief states we could just create an array of beliefs for them which would be pointers to the objects of their awareness, as shown on the slide. But if we do that then our agents will be tracking the truth way too closely. Move the ball from the basket to the box, and the properties of the pointed to objects will change in accordance as shown in the diagram. The red arrow pointing to the Ball will always point to the ball at it’s actual location. If Sally tried to look at what Anne thought, she would still just end up with pointers back to reality. So this won’t do at all.
  • So the next thing to try will be to create for each real object a belief object in the mind of each of the protagonists. We would then have not only Sally, Anne, the Basket, the Box and the Ball, but The Basket for Sally, the Ball for Sally, the Box for Sally, and similarly the Box for Anne, the Basket for Anne, and the Ball for Anne. Perhaps we even duplicate Sally and Anne themselves so that we can keep track of what each believes the other believes. We are certainly closer to solving the problem here, as the diagram shows. But notice that if we add enough agents and have enough objects this could be quite memory intensive, as we would require duplication of every object in the mind of every thinker. If we try to keep track of what each thinks the other thinks, then things get out of hand even faster.

    Furthermore there is a remaining issue: how do we then know that the basket in Sally’s thought is the same basket as the basket in real life, or the basket in Anne’s thought? After all they don’t have the same properties: to wit the Basket in Sally’s thought has a ball in it but not the real basket. So it seems that for the duplicate object solution to work there needs to something the counterpart objects have in common. A good way to do that is to give each object a name. Then it would be clear that Sally and Anne were thinking of the same thing, for example the ball in real life.

    So we then end up with symbolic names for things anyway. With these and relations we have enough and we might as well give up on the duplicated objects. This is what quad stores mentioned earlier permit us to do. If we write this out in N3 - an easy to learn rdf notation - we can see how this enables us to write clearly about people’s beliefs and desires. When we write “:sally believes { :ball in :basket }” we are relating sally to a graph, ie to a set of possible worlds in which the ball is in the basket. Those do not merge with what ann believes, or with what is actually the case such as :ball in box. In the philosophy of language these are known as intensional (with an s) contexts. We can think
    of the first relation relationg :sally to a belief, as relating sally to an object. The content of this object is referentially opaque. We cannot just extract what it contains, unless of course we have reason to believe that it is true. As is well known Truth is a disquotation mechanism.

    Without referential opacity and intensional contexts we would either have to force everyone in the world to write only the truth correctly and never make a mistake, or else we would never be able to fetch information on the web without there being a danger of merging incompatible information and having the equivalent a ball that is in two separate places at once, from which of logically anything would follow.

    The semantic web as it is now understood, gives us the tools to be able to distinguish extensional and intensional contexts, in particular to write out ascriptions of belief. It gives us the tools but it leaves a lot open for further development and investigation
    - There are not yet widely used, or clearly specified propositional attitude relations such as :believes, :desires, ...
    Of course as the semantic web allows for distributed vocabulary creation it is quite easy to create relations for one’s own purposes in order to keep information separated in one belief store.
    - Reasoning engines that track people’s beliefs, such as that Sally will probably be mistaken about the location of the ball because she left the room and did not see it change position, are very very rare. What is easy for a 4 year old child to do is a lot more complex to get computers to do. It is much better to leave it to humans to do that type of reasoning.

  • More practical and feasible is the reverse. Given two descriptions of states of affairs it is quite possible for the computer in the role of the child in the audience to merge information.

    So imagine that instead of the computer trying to guess what Sally and Anne were thinking, they just told the computer what they believed was the case. Perhaps they wrote this out on a web page in rdfa somewhere. The computer could then fetch those pages and place that in its belief store quite confidently. This would be equivalent to Sally and Anne telling the children where each thought the ball was, in a scenario where the children could not see inside the box, and did not know what the truth of the situation was.

    The child in merging those two statements will of course immediately see that that what Anne and Sally say describe two non overlapping sets of possibilities. This is because of what children know of the box and the basket just by looking at them, namely that they are not inside the other, that there is one ball, and that Sally and Anne are speaking of what is the case NOW. Given that, it is clear that there is no way for both girls to be right together. There is no model the child can build that fits what both girls say.

    The computer would be able to come to the same conclusion given enough background information, and an ability to reason with it. This would require information for reasoning on mereological relations such as:
    - the being_inside relation relating two physical objects,
    - that the basket and the box are non overlapping spaces
    - relations on time slice identity, as it is the same moment of a ball that is being spoken about

    Note the importance of the last piece on time slice identity. It is quite possible for one time slice of a ball to be in an earlier time slice of the basket and one time slice of the ball to be in a later time slice of the basket. In plain english the ball could at one time have been in the basket and later in the box. In which case of course both statements would be true. The :in relation then would be quite a general relation relating any object to a 4 dimensional object where a time slice of the first was entirely contained in a time slice of the second. As such we could say that I am :in San Francisco, because I once was there.

    But the children know to focus on the present, when they are asked “where is the ball?”, because the sensual apparatus gives us humans information on the present, and our being in the world is always in the present. It is therefore natural in such conditions to focus on the present as that is also how we interact with the world.

    Relations in the semantic web are very general, and they don’t presuppose such contexts. But let us imagine both that :in was defined in such a way that children could understand it easily. Imagine also that the computer were equipped with the right reasoning engine. It could then reject the merge attempt and point at potential reasons for the impossibility.

    But such reasoning engines are big pieces of machinery. If the semantic web required those it would indeed take a long time to get started, as those machines would need to be widely deployed so that people could understand the functioning and the meaning of those relations.

    Much better therefore is to start with simple relations and to leave the reasoning to the humans. What the computer can do is merge and unmerge information.
  • The web site are an excellent demonstration of how we can design software that allows human intelligence to do what it is best at doing, and make decisions on what should and should not be merged. It also is a very big step forward in semantic web user interface design, as it shows how the user can eaily be made to see the context of where the information he is looking at is coming from.
    I highly recommend looking at the short video on their web site, before going on here. It will help ground the theory we just spoke about.
  • But of course we need to be careful not to fall into a trap of thinking that because of the endless Borgesian Labyrinthine nature of texts that they don’t have a hook into reality at all, or that it is impossible to grasp those. If we look at how we build hyperdata engines, such as the Address Book shown earlier, we notice that we build meaning as algorithms into the software itself. So reasoning engines will understand the purely logical relations and classes, such as owl:sameAs, and owl:inverseFunctionalProperty and know how to use those when merging and deducing information inside of graphs. An address Book will know more about the meaning of the friend of a friend vocabulary..

    On a higher level every node can keep relations retrieved from the web in separate well labelled graphs, which it can then merge on demand, or un merge in case problems are found. It should in more difficult situations ask humans to decide which graphs to trust, or what policies to follow, as that is what we tend to be quite good at deciding.

    From this it follows that every agent will have a somewhat different view on the globally partially incoherent, but locally coherent world wide graph, depending on what information she comes across, what she values and whom she trusts.

    If the web is a database it is going to be a postmodern one, allowing room for the expression of every point of view.
  • For every philosophy there is I have noticed a corresponding Psychology. And where the web is concerned I have noticed the work of Bracha Ettinger a post Lacanian feminist painter/psychoanalyst, whose work I can only say that I have a feeling of understanding. But the themes and the words she uses are very reminiscent of some of the topics we came across. Her Matrixial theory emphasises the web like nature of the subcontious, especially the importance of intersubjectivity. If in our own meanings we must use words coined by other and growing in a web that is mostly independently of us, there is no way we can escape from intersubjectivity, even at the level of meaning. Her emphasis on Borderlinking reminds me of the importance of relations between pages on the web, and how this both may be felt as a fragilisation, as well as in the end a strength.
  • So we are right at the beginning of the Social Web now. The need for it has become clear in the past year, and people are speaking of Personal Data Stores, me of a Social CMS. We are then in a very similar situation where Tim Berners Lee was in 1990 when he showed off the first Web Browser: he was the only one with a some content, and he needed to get the network to grow.

  • To see that these metaphors of growth has a much deeper relevance than one may think, let me summarise this fascinating work by Ruth Garett Millikan, “Language, Thought and Other Biological Categories”. All her papers are online by the way on her web site.
  • R. Millikan starts her book by asking what the distinction is between the physical and the biological. What is the Categorical difference between atoms and life.
  • To start her analysis she considers the well known device: the photocopier. The function of this machine is to make copies of paper. The relation between the original paper and the new one is not a relation of material identity since we end up with two papers. The function of the photocopier is to produce a copy of a certain pattern: the patterns of text on the entry page. “A pattern has been reproduced” she writes “if its form is derived from a previous item or items having, in certain respects the same form, such that had the model(s) been different in these respects, the copy would have been different accordingly”.

    Where it is easy to find the function of a mechanical device such as the photocopier, since we constructed it, and it comes with a user manual, it is not quite so easy to find the proper function for those structures in the world that have arisen by themselves: namely living beings. Since there is no user manual for living beings, which are clearly reproducing themselves Millikan asks how one can find the direct proper functions of parts of an organism such as the heart, the eye, etc... After considering different options she considers the statistical correlation. Perhaps the function of the heart pumping blood is because statistically that is what it does the most. This does indeed sound right in the case of the heart.
  • But the counterexample is just too strong. In the case of sperm and eggs, perhaps the two most fundamental elements in the evolution of most species, the statistical chance of a sperm succeeding in fertilizing an egg is close to zero. This leads Millikan to propose that the primary function of an organism is determined by the evolutionary role that function played in the history of that organism. Since Millikan described the notion of function in terms of the notion of reproduction and evolutionary history, and she places the Biological with the category of that which reproduces, this allows her to extend these thoughts to areas that may have previously seems dissimilar.
    In particular she can then extend these notions to Language. Language can be thought of as composed of a generative syntax which guides the combinations of a finite set of words to produce ever new sentences. The generative syntax in humans can be thought of as being naturally selected for if we believe Chomsky. The words used in a language on the other hand are conventional, and they reproduce between each member of a language community as part of the sentences of which they are part. The function of the words may be many but for many words one important function is the referential function. If I someone shouts “careful! Bus!” the word “Bus” refers to a big heavy vehicle. In a world where people use the word as we do then at least in cases such as this it will have a direct relevance to the life of the person who hears it, as it should trigger them to turn their head and step back. That is the meaning of the words is in part independent of either the speaker or the hearer, though without of course a pattern of use in a community the word would have no meaning at all.
    So with this conception which will remind some of the work on Memes by Richard Dawkins, Millikan has in book made some fundamental moves. First she has naturalised and historicised the philosophy of language. Analytic philosophy had until then suffered from an a-historical because too logical view of the world. By adding evolution, Millikan shows how the etymology of a word is key to understanding its meaning. Secondly she reminds us that there have to be reasons for words to spread from brain to brain. Each person needs to find an advantage in participating in the spreading of the word, which spreading reinforces the use of the word. But very importantly Milikan has also show how the relation of reference between a word and the world can be achieved even if it is not achieved in the minds of every participant that uses the word, as for example many people know something about “HIV” but could not distinguish it from any other disease if asked.
    There may also be here something for people studying Derrida a french philosopher, who wrote a lot about texts to look at. From what I understand from one book by Geoffrey Bennington, Derrida was very aware of the historical nature of words, how when written they were to be read later by many people, many of whome would be unknown to the author, written perhaps to be read after his death. This led Derrida to notice that texts are things that survive over time and whose meaning is partly altered with every reading which in some way escapes them. Indeed if language belongs to the Biological Category, that is the category of those things which follow the logic of reproduction, then Millikan may have found a way for french and analytic philosophy to find another meeting point, perhaps with the added advantage that Millikan can give an explanation also of reference, and so tie the word back to the world.
  • So with such an understanding of language we can explain the final part of how the Social Web can grow in a distributed decentralised way. In a decentralised World Wide Web there can be no Académie du Web that orders and defines the meanings of URLs on the Web. Such an organisation would be a huge bottleneck, and it would be impossible in a global space. But if there is no central authority, only relative trusted authorities such as the W3C, then how do we avoid a massive duplication of vocabularies on every subject? Who gets to decide which the good vocabularies are? Well in the end I want to suggest it is simply natural selection of vocabularies. This natural selection occurs in a number of ways, which are not that different from how HTML and the web got going in the first place. Vocabularies appear in documents that are read by software, which can then publish new documents for others to link to (in the case of the WWW the navigators did not publish, but html was easy enough for people publishers to learn by themselves). If doing this increases the value to the user of the software tool he is using while increasing the value of the network even more then the vocabulary (the URIs, html) will spread in the web and so will the software too. As these networks grow in value, any new software tool that consumes the vocabulary will find it more advantageous to work with the old vocabulary rather than create a new one to do the same thing, as it will then surf along with the growing movement that already exists. For a vocabulary to maximise it’s adoption it has to minimise any friction to its duplication: patent free, copyright free licences help a lot here, and so does truthfulness and trust in the institutions publishing core vocabularies - a place for universities, which are some of the longest established institutions in the world, to play a leading roles.
  • So in conclusion of this very long talk. We saw how behind the web, in the social web and in the linked data and semantic web the network effect is a driving force that allows these distributed peer to peer, biological like systems to grow and stabilise. Currently these network effects are working in favour of some very big providers, but their inability to link out is imprisoning their users and in the longer term themselves too. A much wider space of possibilities is available in a global social web allowing players from health agents to government institutions to political players to businesses to work together in a decentralised way, protecting each actors privacy in the process, as well as allowing them each to express their point of view without enforcing a point of view on the system as a whole. It is on the foundations of the URI, the global identifier, that the network effect can take place, by tying sense and reference together and enabling information linking and merging at a global scale.
  • Identity is in a very strong way social, and of course the social is very much tied to identity. This should not come as a surprise. There is a lot more to be said certainly but the best is to leave this open for further exploration and discussion on the web.